GPT-7 Release Date: What We Know, and Why the Next GPT Probably Isn't GPT-7

GPT-7 is not out, and OpenAI has never announced it. OpenAI’s newest model is GPT-6.1 Terra, released on 26 October 2026, and its most capable public model is still GPT-6 Astra, released on 3 September 2026. On 2 November, Sam Altman said OpenAI’s “next GPT” is coming “soon”. Most of the evidence says that model will be a GPT-6.2, not GPT-7.

That one phrase has set off a new round of GPT-7 posts, on top of a leaked codename, a prediction market and a viral Altman clip. This guide checks each claim against OpenAI’s own pages, explains what Altman actually said, and gives the most likely GPT-7 release window with the reasoning behind it.

The Key Takeaways

• Not released: OpenAI’s model docs and API price list name no GPT-7. The top models are GPT-6 Astra, GPT-6.1 Sol and GPT-6.1 Terra.
• The “next GPT”: Altman said on 2 November that the model behind OpenAI’s Navier-Stokes proof will be released soon. Altman did not say GPT-7, and OpenAI has not named the model.
• GPT-6.2 sightings: a gpt-6.2-astra model id has shown up in Codex usage stats, and The Information reported new Astra and Sol models as soon as 6 November.
• Timing: Polymarket traders give a public GPT-7 a 49% chance by 30 June 2027 and 87% by 31 December 2027.
• The Bel leak: the claim of a 10-trillion-parameter pretrain called Bel comes from one anonymous X account. OpenAI has never confirmed it.

Is GPT-7 Out?

No. GPT-7 has not been released, previewed or named by OpenAI. OpenAI’s ChatGPT and Codex model guide, read on 5 November 2026, recommends four models: GPT-6 Astra, GPT-6.1 Sol, GPT-6.1 Terra and GPT-6 Luna. OpenAI’s API price list has rows for the same models and no row for GPT-7 or GPT-6.2.

The table shows OpenAI’s current lineup and where GPT-7 stands. Prices are OpenAI’s standard API rates per million tokens.

Much of the “GPT-7 is out” talk comes from people mixing up generations. A GPT-6 Astra answer is not a GPT-7 answer, and no ChatGPT plan offers a model called GPT-7. Our ChatGPT 6 release guide covers how the GPT-6 family reached each plan.

When Will GPT-7 Come Out?

OpenAI has given no date for GPT-7. The most likely window is the second half of 2027, based on OpenAI’s release gaps and on prediction-market odds. Neither is a promise from OpenAI, so each input is shown separately.

OpenAI’s Release Gaps Are Shrinking

OpenAI’s whole-number GPT releases are coming faster. The gap from GPT-4 to GPT-5 was about 29 months. The gap from GPT-5 to GPT-6 Astra was about 13 months.

A repeat of the 13-month gap puts GPT-7 around October 2027. A shorter gap would bring it forward into the first half of 2027. Point releases fill the time in between. OpenAI shipped GPT-5.5 in April 2026, GPT-5.6 Sol in July and five GPT-6 models between 3 September and 26 October.

What Polymarket Traders Expect

Polymarket traders expect GPT-7 in 2027. The “GPT-7 released by...?” market opened on 4 September 2026, the day after Astra launched. On 5 November 2026 it priced a public GPT-7 at 49% by 30 June 2027 and 87% by 31 December 2027. Total volume was $24,751, which is small, so a few trades can move these numbers.

The market’s rules matter more than the odds. A model only counts if its name or model id includes “GPT” and a version of 7 or higher. A GPT-7 Astra or GPT-7 Sol would count. A GPT-6.2 would not, however powerful it is, and neither would a Codex or cyber model. A closed beta does not count either: the model must be open to the general public.

Is Altman’s “Next GPT” GPT-7?

Probably not. Altman’s “next GPT” is the internal model that produced OpenAI’s Navier-Stokes proof in September, and the hardest evidence about its name points to GPT-6.2. OpenAI has not named it, so this section lays out what each piece of evidence says.

What Altman Said in Abu Dhabi

Sam Altman spoke at the ADIPEC energy conference in Abu Dhabi on 2 November 2026. Describing OpenAI’s first Millennium Prize proof, he said: “It was 10,000 agents of the model that is going to be released soon, the next GPT, working for about four days, nonstop.” He gave no name, no date and no price. This post carried the clip and spread the line:

Sam Altman says that the internal model that solved Navier-Stokes will be released "soon":

"It was 10,000 agents of the model that is going to be released soon, the next GPT, working for about four days, nonstop, collaborating with each other perfectly."
https://youtu.be/2YWfVGpVtvc?t=332

— prinz (@deredleritt3r) November 2, 2026

“The next GPT” is the phrase that turned into GPT-7 headlines. Altman did not say GPT-7. He said the next GPT, and OpenAI has called every release in the GPT-6 family a GPT.

What OpenAI Wrote About the Navier-Stokes Model

OpenAI described the Navier-Stokes model in its own words on 8 September 2026. Its official account called it “an OpenAI next-generation model significantly more capable than GPT-6 Astra”:

We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics.

The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra.

The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.

— OpenAI (@OpenAI) September 8, 2026

The full Navier-Stokes post adds the detail that matters for GPT-7. OpenAI says it had been training the model since 28 August and that training was “ongoing”. It also says the model “was developed through large-scale reinforcement learning on top of a previously pretrained model”.

That sentence settles less than it seems. It says the model is an existing base with more training on top, not a pretrain started in September. It does not say which base. If the base is the rumored Bel pretrain, the next GPT could be the first model of a new generation. The name OpenAI uses in its own systems is a better guide. The proof itself is covered in our guide to math problems AI solved in 2026.

Why the Evidence Points to GPT-6.2

Three separate signals point to a GPT-6.2 as OpenAI’s next big release. None of them is an announcement, but they agree with each other.

The first is a model id. TestingCatalog reported on 2 November that several Codex users saw gpt-6.2-astra in their usage stats on 31 October, while the model picker still said GPT-6 Astra. That pattern looks like quiet testing on real traffic.

The second is a news report. The Information reported that OpenAI could release next-generation Astra and Sol models and a cybersecurity model as soon as Friday, 6 November. PANews and ChainCatcher relayed the story on 5 November. The report did not give model numbers.

The third is the gap in OpenAI’s lineup. OpenAI planned a GPT-6.1 Astra for October and cancelled it after safety testing. The Wall Street Journal reported that the model “regressed in two areas”: honesty about its own actions and asking users for permission. Our GPT-6.1 Sol guide covers that decision. Sol and Terra have both reached version 6.1 since then, and Astra has not, so a new Astra is the obvious next step.

If OpenAI does call its next model GPT-7, this reading is wrong, and nothing OpenAI has published rules that out. A model id seen by several Codex users is still the strongest clue to the name.

What Sam Altman Has Actually Said About GPT-7

Sam Altman has mentioned GPT-7 only in passing, and never as a product with a date. Three quotes do most of the work in GPT-7 search results, and one widely repeated quote could not be found at all.

The Cheeky Pint Remark

Altman’s most recent GPT-7 mention came on 29 October 2026, on Stripe co-founder John Collison’s Cheeky Pint podcast. Collison said “we should expect GPT-7 to be really good now”, because OpenAI can use GPT-6 to help build it. Altman answered with a question OpenAI staff debate: “What are the questions that you’re supposed to ask of GPT-7?” That is a remark about AI doing AI research. It is not a release hint.

The “ChatGPT 7” Clip Is From 2025

The most viral GPT-7 quote is a clip in which Altman imagines a president handing decisions to “ChatGPT 7”. It spread again on 13 October 2026 and passed half a million views:

Sam Altman: What happens when AI gets so smart that even the President can’t do better than following its recommendations?

“What if the President cannot do better than following ChatGPT 7’s recommendation, but can’t really understand it either?”

“What if I just say, ‘ChatGPT 7, you are in charge. Good luck.’”

“Society has collectively transitioned a significant part of decision-making to this very powerful system.”

“That’s how I think things can go wrong.”

— vik (@vikktorrrre) October 13, 2026

The clip is old. Altman made those remarks at a Federal Reserve banking conference in Washington in July 2025, when The Guardian reported his warning that presidents might follow ChatGPT’s recommendations. “ChatGPT 7” was a stand-in for a future model, not a roadmap item.

The GPT-6-7 Joke and the Missing $0 Quote

Altman has also joked about the name. On 30 October 2025 he posted on X that “GPT-6 will be renamed GPT-6-7, you’re welcome”, a nod to the “6-7” meme that Windows Central and others picked up. OpenAI shipped GPT-6 under its normal name.

One GPT-7 quote we could not verify at all. Several GPT-7 pages say Altman posted in December 2023 that “someday you’ll get $0 GPT-7”. We found no copy of that post. Altman’s real post that day said “feeling good about this prediction!” about his 2021 forecast that the costs of intelligence and energy are heading toward near-zero. It does not mention GPT-7.

GPT-7 Rumors and Leaks, Graded

No GPT-7 leak has been confirmed. The biggest rumor, a pretrain called Bel, traces back to one anonymous account. Several newer claims are about GPT-6.2 and got relabelled as GPT-7 along the way.

The Bel Leak

The Bel leak is a claim about an OpenAI pretraining run, not a product. On 25 August 2026, an anonymous X account called @synthwavedd claimed OpenAI had finished pretraining a model called Bel. The post called Bel the successor to an earlier run called Doug, with more than 10 trillion total parameters. The claim said Bel was meant as the base for models after GPT-6. The post is still up, with more than 800,000 views, and every report on Bel that we found cites it.

Two points keep Bel in proportion. A pretrained base is raw material: it needs months of further training and safety testing before it ships. And a total parameter count says little about cost or speed in a mixture-of-experts model, where only part of the model runs for each token. Searches for “GPT 7 Bel” and “GPT 7 Bell” are about this leak. OpenAI has never used either name.

GPT-6.2 Betelgeuse

GPT-6.2 Betelgeuse is a name from a single screenshot. Around 28 October, a Discord bot in an unofficial “internal tracker” server listed a model string gpt-6.2-betelgeuse, according to OrcaRouter’s write-up of the thread. Betelgeuse is a red supergiant star whose name starts with “Bel”, so some posters decided Bel stands for Betelgeuse. Nobody has shown the string in OpenAI’s apps or API, and the screenshot’s source is unknown.

The name fits OpenAI’s habit of naming tiers after stars and planets: Sol, Luna, Terra and Astra. That makes it plausible. It also makes it easy to invent.

Every GPT-7 Claim, Checked

The table grades every GPT-7 claim we found against OpenAI’s own pages and the original posts.

GPT-7 Compute and the “7 Trillion” Searches

Searches for “ChatGPT 7 trillion” are not about a GPT-7 spec. They come from a February 2024 Wall Street Journal report that Altman was seeking as much as $7 trillion to expand the world’s chip-making capacity, as Al Jazeera summarised it. The number was dollars for chip plants, not parameters in a model.

Compute is still the real limit on GPT-7. Altman put the cost of the Navier-Stokes run alone at “order $10 million” on the Cheeky Pint podcast, and that run used a model OpenAI had already trained. A new, larger base model costs far more to build than any single run on top of it. That is one reason whole-number GPT releases still come more than a year apart.

What Will GPT-7 Be Like?

Nobody outside OpenAI knows what GPT-7 will be like, and OpenAI has published nothing about it. Three things OpenAI has said or done recently are the best guide to what any model after GPT-6 Astra will look like.

It will probably be priced per token. That is a safe bet for OpenAI’s next model, and likely for GPT-7 after it. Altman told Cheeky Pint that OpenAI “almost” decided not to price its next model in tokens at all, and dropped the idea after a negative customer response.

It will probably arrive in stages. GPT-6 Astra went to a small set of organisations first and reached paying ChatGPT users the next day. Free and Go users never got Astra in Chat, as our ChatGPT pricing guide shows. A free GPT-7 at launch would break that pattern.

It will probably be built for long agent work. The Navier-Stokes run used about 10,000 agents for about 88 hours, by OpenAI’s own account. That is the use Altman keeps returning to: many copies of one model working together on a single hard problem.

How You Will Know GPT-7 Is Real

A real GPT-7 will show up in OpenAI’s own documents before anywhere else. Four first-party signals separate a launch from a rumor.

A model page and an API id. Every GPT-6 model so far has appeared in OpenAI’s model guide and API price list with a rate per million tokens. A GPT-7 without a price is not a launch.

A system card. OpenAI rated GPT-6 Astra its first “Critical” model for cybersecurity under its Preparedness Framework. Any stronger model needs that evaluation, and OpenAI publishes it.

A dated release note. The ChatGPT release notes name each model and the plans that get it.

A post from @OpenAI. Screenshots, Discord bots and countdown accounts do not count, however convincing they look.

What to Use While You Wait for GPT-7

GPT-6 Astra is the strongest OpenAI model you can use today. In ChatGPT it powers the Pro reasoning option on the Pro, Business and Enterprise plans, and Plus users get it in ChatGPT Work and Codex. Developers can call it in the API at $10 per million input tokens and $50 per million output tokens. Our GPT-6 Astra guide covers what it does well and where it falls short.

For most work, GPT-6.1 Sol and GPT-6.1 Terra are the better value. GPT-6.1 Sol costs a fifth of Astra’s API rate, and OpenAI says GPT-6.1 Terra matches Astra on its computer-use benchmark at a tenth of the price per token. If a GPT-6.2 lands this week, it will most likely sit at the top of the same lineup. Fello AI lets you try each new OpenAI model without a new account. New models usually arrive in the app as soon as their API opens, next to Claude and Gemini in the same window.

The Bottom Line

GPT-7 is a 2027 story. The model OpenAI is about to release is the one that solved Navier-Stokes, and the only name seen in OpenAI’s own systems is gpt-6.2-astra. A model id in Codex is harder evidence than a phrase from a conference stage. Treat any GPT-7 headline this month as a GPT-6.2 headline until OpenAI’s model guide says otherwise, and plan around the second half of 2027 for the real thing. Google’s next generation runs on a similar clock: our Gemini 5 release date guide estimates it between July and October 2027. Anthropic is likely to move sooner: our Claude 6 release date guide puts the first public Claude 6 model in the first half of 2027.

FAQ

Das Titelbild dieses Artikels wurde mithilfe von KI erstellt.